1/* ------------------------------------------------------------
2 * ibmvscsi.c
3 * (C) Copyright IBM Corporation 1994, 2004
4 * Authors: Colin DeVilbiss (devilbis@us.ibm.com)
5 *          Santiago Leon (santil@us.ibm.com)
6 *          Dave Boutcher (sleddog@us.ibm.com)
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307
21 * USA
22 *
23 * ------------------------------------------------------------
24 * Emulation of a SCSI host adapter for Virtual I/O devices
25 *
26 * This driver supports the SCSI adapter implemented by the IBM
27 * Power5 firmware.  That SCSI adapter is not a physical adapter,
28 * but allows Linux SCSI peripheral drivers to directly
29 * access devices in another logical partition on the physical system.
30 *
31 * The virtual adapter(s) are present in the open firmware device
32 * tree just like real adapters.
33 *
34 * One of the capabilities provided on these systems is the ability
35 * to DMA between partitions.  The architecture states that for VSCSI,
36 * the server side is allowed to DMA to and from the client.  The client
37 * is never trusted to DMA to or from the server directly.
38 *
39 * Messages are sent between partitions on a "Command/Response Queue"
40 * (CRQ), which is just a buffer of 16 byte entries in the receiver's
41 * Senders cannot access the buffer directly, but send messages by
42 * making a hypervisor call and passing in the 16 bytes.  The hypervisor
43 * puts the message in the next 16 byte space in round-robin fashion,
44 * turns on the high order bit of the message (the valid bit), and
45 * generates an interrupt to the receiver (if interrupts are turned on.)
46 * The receiver just turns off the valid bit when they have copied out
47 * the message.
48 *
49 * The VSCSI client builds a SCSI Remote Protocol (SRP) Information Unit
50 * (IU) (as defined in the T10 standard available at www.t10.org), gets
51 * a DMA address for the message, and sends it to the server as the
52 * payload of a CRQ message.  The server DMAs the SRP IU and processes it,
53 * including doing any additional data transfers.  When it is done, it
54 * DMAs the SRP response back to the same address as the request came from,
55 * and sends a CRQ message back to inform the client that the request has
56 * completed.
57 *
58 * TODO: This is currently pretty tied to the IBM pSeries hypervisor
59 * interfaces.  It would be really nice to abstract this above an RDMA
60 * layer.
61 */
62
63#include <linux/module.h>
64#include <linux/moduleparam.h>
65#include <linux/dma-mapping.h>
66#include <linux/delay.h>
67#include <linux/slab.h>
68#include <linux/of.h>
69#include <linux/pm.h>
70#include <linux/kthread.h>
71#include <asm/firmware.h>
72#include <asm/vio.h>
73#include <scsi/scsi.h>
74#include <scsi/scsi_cmnd.h>
75#include <scsi/scsi_host.h>
76#include <scsi/scsi_device.h>
77#include <scsi/scsi_transport_srp.h>
78#include "ibmvscsi.h"
79
80/* The values below are somewhat arbitrary default values, but
81 * OS/400 will use 3 busses (disks, CDs, tapes, I think.)
82 * Note that there are 3 bits of channel value, 6 bits of id, and
83 * 5 bits of LUN.
84 */
85static int max_id = 64;
86static int max_channel = 3;
87static int init_timeout = 300;
88static int login_timeout = 60;
89static int info_timeout = 30;
90static int abort_timeout = 60;
91static int reset_timeout = 60;
92static int max_requests = IBMVSCSI_MAX_REQUESTS_DEFAULT;
93static int max_events = IBMVSCSI_MAX_REQUESTS_DEFAULT + 2;
94static int fast_fail = 1;
95static int client_reserve = 1;
96static char partition_name[97] = "UNKNOWN";
97static unsigned int partition_number = -1;
98
99static struct scsi_transport_template *ibmvscsi_transport_template;
100
101#define IBMVSCSI_VERSION "1.5.9"
102
103MODULE_DESCRIPTION("IBM Virtual SCSI");
104MODULE_AUTHOR("Dave Boutcher");
105MODULE_LICENSE("GPL");
106MODULE_VERSION(IBMVSCSI_VERSION);
107
108module_param_named(max_id, max_id, int, S_IRUGO | S_IWUSR);
109MODULE_PARM_DESC(max_id, "Largest ID value for each channel");
110module_param_named(max_channel, max_channel, int, S_IRUGO | S_IWUSR);
111MODULE_PARM_DESC(max_channel, "Largest channel value");
112module_param_named(init_timeout, init_timeout, int, S_IRUGO | S_IWUSR);
113MODULE_PARM_DESC(init_timeout, "Initialization timeout in seconds");
114module_param_named(max_requests, max_requests, int, S_IRUGO);
115MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter");
116module_param_named(fast_fail, fast_fail, int, S_IRUGO | S_IWUSR);
117MODULE_PARM_DESC(fast_fail, "Enable fast fail. [Default=1]");
118module_param_named(client_reserve, client_reserve, int, S_IRUGO );
119MODULE_PARM_DESC(client_reserve, "Attempt client managed reserve/release");
120
121static void ibmvscsi_handle_crq(struct viosrp_crq *crq,
122				struct ibmvscsi_host_data *hostdata);
123
124/* ------------------------------------------------------------
125 * Routines for managing the command/response queue
126 */
127/**
128 * ibmvscsi_handle_event: - Interrupt handler for crq events
129 * @irq:	number of irq to handle, not used
130 * @dev_instance: ibmvscsi_host_data of host that received interrupt
131 *
132 * Disables interrupts and schedules srp_task
133 * Always returns IRQ_HANDLED
134 */
135static irqreturn_t ibmvscsi_handle_event(int irq, void *dev_instance)
136{
137	struct ibmvscsi_host_data *hostdata =
138	    (struct ibmvscsi_host_data *)dev_instance;
139	vio_disable_interrupts(to_vio_dev(hostdata->dev));
140	tasklet_schedule(&hostdata->srp_task);
141	return IRQ_HANDLED;
142}
143
144/**
145 * release_crq_queue: - Deallocates data and unregisters CRQ
146 * @queue:	crq_queue to initialize and register
147 * @host_data:	ibmvscsi_host_data of host
148 *
149 * Frees irq, deallocates a page for messages, unmaps dma, and unregisters
150 * the crq with the hypervisor.
151 */
152static void ibmvscsi_release_crq_queue(struct crq_queue *queue,
153				       struct ibmvscsi_host_data *hostdata,
154				       int max_requests)
155{
156	long rc = 0;
157	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
158	free_irq(vdev->irq, (void *)hostdata);
159	tasklet_kill(&hostdata->srp_task);
160	do {
161		if (rc)
162			msleep(100);
163		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
164	} while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
165	dma_unmap_single(hostdata->dev,
166			 queue->msg_token,
167			 queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
168	free_page((unsigned long)queue->msgs);
169}
170
171/**
172 * crq_queue_next_crq: - Returns the next entry in message queue
173 * @queue:	crq_queue to use
174 *
175 * Returns pointer to next entry in queue, or NULL if there are no new
176 * entried in the CRQ.
177 */
178static struct viosrp_crq *crq_queue_next_crq(struct crq_queue *queue)
179{
180	struct viosrp_crq *crq;
181	unsigned long flags;
182
183	spin_lock_irqsave(&queue->lock, flags);
184	crq = &queue->msgs[queue->cur];
185	if (crq->valid & 0x80) {
186		if (++queue->cur == queue->size)
187			queue->cur = 0;
188
189		/* Ensure the read of the valid bit occurs before reading any
190		 * other bits of the CRQ entry
191		 */
192		rmb();
193	} else
194		crq = NULL;
195	spin_unlock_irqrestore(&queue->lock, flags);
196
197	return crq;
198}
199
200/**
201 * ibmvscsi_send_crq: - Send a CRQ
202 * @hostdata:	the adapter
203 * @word1:	the first 64 bits of the data
204 * @word2:	the second 64 bits of the data
205 */
206static int ibmvscsi_send_crq(struct ibmvscsi_host_data *hostdata,
207			     u64 word1, u64 word2)
208{
209	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
210
211	/*
212	 * Ensure the command buffer is flushed to memory before handing it
213	 * over to the VIOS to prevent it from fetching any stale data.
214	 */
215	mb();
216	return plpar_hcall_norets(H_SEND_CRQ, vdev->unit_address, word1, word2);
217}
218
219/**
220 * ibmvscsi_task: - Process srps asynchronously
221 * @data:	ibmvscsi_host_data of host
222 */
223static void ibmvscsi_task(void *data)
224{
225	struct ibmvscsi_host_data *hostdata = (struct ibmvscsi_host_data *)data;
226	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
227	struct viosrp_crq *crq;
228	int done = 0;
229
230	while (!done) {
231		/* Pull all the valid messages off the CRQ */
232		while ((crq = crq_queue_next_crq(&hostdata->queue)) != NULL) {
233			ibmvscsi_handle_crq(crq, hostdata);
234			crq->valid = 0x00;
235		}
236
237		vio_enable_interrupts(vdev);
238		crq = crq_queue_next_crq(&hostdata->queue);
239		if (crq != NULL) {
240			vio_disable_interrupts(vdev);
241			ibmvscsi_handle_crq(crq, hostdata);
242			crq->valid = 0x00;
243		} else {
244			done = 1;
245		}
246	}
247}
248
249static void gather_partition_info(void)
250{
251	struct device_node *rootdn;
252
253	const char *ppartition_name;
254	const __be32 *p_number_ptr;
255
256	/* Retrieve information about this partition */
257	rootdn = of_find_node_by_path("/");
258	if (!rootdn) {
259		return;
260	}
261
262	ppartition_name = of_get_property(rootdn, "ibm,partition-name", NULL);
263	if (ppartition_name)
264		strncpy(partition_name, ppartition_name,
265				sizeof(partition_name));
266	p_number_ptr = of_get_property(rootdn, "ibm,partition-no", NULL);
267	if (p_number_ptr)
268		partition_number = of_read_number(p_number_ptr, 1);
269	of_node_put(rootdn);
270}
271
272static void set_adapter_info(struct ibmvscsi_host_data *hostdata)
273{
274	memset(&hostdata->madapter_info, 0x00,
275			sizeof(hostdata->madapter_info));
276
277	dev_info(hostdata->dev, "SRP_VERSION: %s\n", SRP_VERSION);
278	strcpy(hostdata->madapter_info.srp_version, SRP_VERSION);
279
280	strncpy(hostdata->madapter_info.partition_name, partition_name,
281			sizeof(hostdata->madapter_info.partition_name));
282
283	hostdata->madapter_info.partition_number =
284					cpu_to_be32(partition_number);
285
286	hostdata->madapter_info.mad_version = cpu_to_be32(1);
287	hostdata->madapter_info.os_type = cpu_to_be32(2);
288}
289
290/**
291 * reset_crq_queue: - resets a crq after a failure
292 * @queue:	crq_queue to initialize and register
293 * @hostdata:	ibmvscsi_host_data of host
294 *
295 */
296static int ibmvscsi_reset_crq_queue(struct crq_queue *queue,
297				    struct ibmvscsi_host_data *hostdata)
298{
299	int rc = 0;
300	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
301
302	/* Close the CRQ */
303	do {
304		if (rc)
305			msleep(100);
306		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
307	} while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
308
309	/* Clean out the queue */
310	memset(queue->msgs, 0x00, PAGE_SIZE);
311	queue->cur = 0;
312
313	set_adapter_info(hostdata);
314
315	/* And re-open it again */
316	rc = plpar_hcall_norets(H_REG_CRQ,
317				vdev->unit_address,
318				queue->msg_token, PAGE_SIZE);
319	if (rc == 2) {
320		/* Adapter is good, but other end is not ready */
321		dev_warn(hostdata->dev, "Partner adapter not ready\n");
322	} else if (rc != 0) {
323		dev_warn(hostdata->dev, "couldn't register crq--rc 0x%x\n", rc);
324	}
325	return rc;
326}
327
328/**
329 * initialize_crq_queue: - Initializes and registers CRQ with hypervisor
330 * @queue:	crq_queue to initialize and register
331 * @hostdata:	ibmvscsi_host_data of host
332 *
333 * Allocates a page for messages, maps it for dma, and registers
334 * the crq with the hypervisor.
335 * Returns zero on success.
336 */
337static int ibmvscsi_init_crq_queue(struct crq_queue *queue,
338				   struct ibmvscsi_host_data *hostdata,
339				   int max_requests)
340{
341	int rc;
342	int retrc;
343	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
344
345	queue->msgs = (struct viosrp_crq *)get_zeroed_page(GFP_KERNEL);
346
347	if (!queue->msgs)
348		goto malloc_failed;
349	queue->size = PAGE_SIZE / sizeof(*queue->msgs);
350
351	queue->msg_token = dma_map_single(hostdata->dev, queue->msgs,
352					  queue->size * sizeof(*queue->msgs),
353					  DMA_BIDIRECTIONAL);
354
355	if (dma_mapping_error(hostdata->dev, queue->msg_token))
356		goto map_failed;
357
358	gather_partition_info();
359	set_adapter_info(hostdata);
360
361	retrc = rc = plpar_hcall_norets(H_REG_CRQ,
362				vdev->unit_address,
363				queue->msg_token, PAGE_SIZE);
364	if (rc == H_RESOURCE)
365		/* maybe kexecing and resource is busy. try a reset */
366		rc = ibmvscsi_reset_crq_queue(queue,
367					      hostdata);
368
369	if (rc == 2) {
370		/* Adapter is good, but other end is not ready */
371		dev_warn(hostdata->dev, "Partner adapter not ready\n");
372		retrc = 0;
373	} else if (rc != 0) {
374		dev_warn(hostdata->dev, "Error %d opening adapter\n", rc);
375		goto reg_crq_failed;
376	}
377
378	queue->cur = 0;
379	spin_lock_init(&queue->lock);
380
381	tasklet_init(&hostdata->srp_task, (void *)ibmvscsi_task,
382		     (unsigned long)hostdata);
383
384	if (request_irq(vdev->irq,
385			ibmvscsi_handle_event,
386			0, "ibmvscsi", (void *)hostdata) != 0) {
387		dev_err(hostdata->dev, "couldn't register irq 0x%x\n",
388			vdev->irq);
389		goto req_irq_failed;
390	}
391
392	rc = vio_enable_interrupts(vdev);
393	if (rc != 0) {
394		dev_err(hostdata->dev, "Error %d enabling interrupts!!!\n", rc);
395		goto req_irq_failed;
396	}
397
398	return retrc;
399
400      req_irq_failed:
401	tasklet_kill(&hostdata->srp_task);
402	rc = 0;
403	do {
404		if (rc)
405			msleep(100);
406		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
407	} while ((rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
408      reg_crq_failed:
409	dma_unmap_single(hostdata->dev,
410			 queue->msg_token,
411			 queue->size * sizeof(*queue->msgs), DMA_BIDIRECTIONAL);
412      map_failed:
413	free_page((unsigned long)queue->msgs);
414      malloc_failed:
415	return -1;
416}
417
418/**
419 * reenable_crq_queue: - reenables a crq after
420 * @queue:	crq_queue to initialize and register
421 * @hostdata:	ibmvscsi_host_data of host
422 *
423 */
424static int ibmvscsi_reenable_crq_queue(struct crq_queue *queue,
425				       struct ibmvscsi_host_data *hostdata)
426{
427	int rc = 0;
428	struct vio_dev *vdev = to_vio_dev(hostdata->dev);
429
430	/* Re-enable the CRQ */
431	do {
432		if (rc)
433			msleep(100);
434		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
435	} while ((rc == H_IN_PROGRESS) || (rc == H_BUSY) || (H_IS_LONG_BUSY(rc)));
436
437	if (rc)
438		dev_err(hostdata->dev, "Error %d enabling adapter\n", rc);
439	return rc;
440}
441
442/* ------------------------------------------------------------
443 * Routines for the event pool and event structs
444 */
445/**
446 * initialize_event_pool: - Allocates and initializes the event pool for a host
447 * @pool:	event_pool to be initialized
448 * @size:	Number of events in pool
449 * @hostdata:	ibmvscsi_host_data who owns the event pool
450 *
451 * Returns zero on success.
452*/
453static int initialize_event_pool(struct event_pool *pool,
454				 int size, struct ibmvscsi_host_data *hostdata)
455{
456	int i;
457
458	pool->size = size;
459	pool->next = 0;
460	pool->events = kcalloc(pool->size, sizeof(*pool->events), GFP_KERNEL);
461	if (!pool->events)
462		return -ENOMEM;
463
464	pool->iu_storage =
465	    dma_alloc_coherent(hostdata->dev,
466			       pool->size * sizeof(*pool->iu_storage),
467			       &pool->iu_token, 0);
468	if (!pool->iu_storage) {
469		kfree(pool->events);
470		return -ENOMEM;
471	}
472
473	for (i = 0; i < pool->size; ++i) {
474		struct srp_event_struct *evt = &pool->events[i];
475		memset(&evt->crq, 0x00, sizeof(evt->crq));
476		atomic_set(&evt->free, 1);
477		evt->crq.valid = 0x80;
478		evt->crq.IU_length = cpu_to_be16(sizeof(*evt->xfer_iu));
479		evt->crq.IU_data_ptr = cpu_to_be64(pool->iu_token +
480			sizeof(*evt->xfer_iu) * i);
481		evt->xfer_iu = pool->iu_storage + i;
482		evt->hostdata = hostdata;
483		evt->ext_list = NULL;
484		evt->ext_list_token = 0;
485	}
486
487	return 0;
488}
489
490/**
491 * release_event_pool: - Frees memory of an event pool of a host
492 * @pool:	event_pool to be released
493 * @hostdata:	ibmvscsi_host_data who owns the even pool
494 *
495 * Returns zero on success.
496*/
497static void release_event_pool(struct event_pool *pool,
498			       struct ibmvscsi_host_data *hostdata)
499{
500	int i, in_use = 0;
501	for (i = 0; i < pool->size; ++i) {
502		if (atomic_read(&pool->events[i].free) != 1)
503			++in_use;
504		if (pool->events[i].ext_list) {
505			dma_free_coherent(hostdata->dev,
506				  SG_ALL * sizeof(struct srp_direct_buf),
507				  pool->events[i].ext_list,
508				  pool->events[i].ext_list_token);
509		}
510	}
511	if (in_use)
512		dev_warn(hostdata->dev, "releasing event pool with %d "
513			 "events still in use?\n", in_use);
514	kfree(pool->events);
515	dma_free_coherent(hostdata->dev,
516			  pool->size * sizeof(*pool->iu_storage),
517			  pool->iu_storage, pool->iu_token);
518}
519
520/**
521 * valid_event_struct: - Determines if event is valid.
522 * @pool:	event_pool that contains the event
523 * @evt:	srp_event_struct to be checked for validity
524 *
525 * Returns zero if event is invalid, one otherwise.
526*/
527static int valid_event_struct(struct event_pool *pool,
528				struct srp_event_struct *evt)
529{
530	int index = evt - pool->events;
531	if (index < 0 || index >= pool->size)	/* outside of bounds */
532		return 0;
533	if (evt != pool->events + index)	/* unaligned */
534		return 0;
535	return 1;
536}
537
538/**
539 * ibmvscsi_free-event_struct: - Changes status of event to "free"
540 * @pool:	event_pool that contains the event
541 * @evt:	srp_event_struct to be modified
542 *
543*/
544static void free_event_struct(struct event_pool *pool,
545				       struct srp_event_struct *evt)
546{
547	if (!valid_event_struct(pool, evt)) {
548		dev_err(evt->hostdata->dev, "Freeing invalid event_struct %p "
549			"(not in pool %p)\n", evt, pool->events);
550		return;
551	}
552	if (atomic_inc_return(&evt->free) != 1) {
553		dev_err(evt->hostdata->dev, "Freeing event_struct %p "
554			"which is not in use!\n", evt);
555		return;
556	}
557}
558
559/**
560 * get_evt_struct: - Gets the next free event in pool
561 * @pool:	event_pool that contains the events to be searched
562 *
563 * Returns the next event in "free" state, and NULL if none are free.
564 * Note that no synchronization is done here, we assume the host_lock
565 * will syncrhonze things.
566*/
567static struct srp_event_struct *get_event_struct(struct event_pool *pool)
568{
569	int i;
570	int poolsize = pool->size;
571	int offset = pool->next;
572
573	for (i = 0; i < poolsize; i++) {
574		offset = (offset + 1) % poolsize;
575		if (!atomic_dec_if_positive(&pool->events[offset].free)) {
576			pool->next = offset;
577			return &pool->events[offset];
578		}
579	}
580
581	printk(KERN_ERR "ibmvscsi: found no event struct in pool!\n");
582	return NULL;
583}
584
585/**
586 * init_event_struct: Initialize fields in an event struct that are always
587 *                    required.
588 * @evt:        The event
589 * @done:       Routine to call when the event is responded to
590 * @format:     SRP or MAD format
591 * @timeout:    timeout value set in the CRQ
592 */
593static void init_event_struct(struct srp_event_struct *evt_struct,
594			      void (*done) (struct srp_event_struct *),
595			      u8 format,
596			      int timeout)
597{
598	evt_struct->cmnd = NULL;
599	evt_struct->cmnd_done = NULL;
600	evt_struct->sync_srp = NULL;
601	evt_struct->crq.format = format;
602	evt_struct->crq.timeout = cpu_to_be16(timeout);
603	evt_struct->done = done;
604}
605
606/* ------------------------------------------------------------
607 * Routines for receiving SCSI responses from the hosting partition
608 */
609
610/**
611 * set_srp_direction: Set the fields in the srp related to data
612 *     direction and number of buffers based on the direction in
613 *     the scsi_cmnd and the number of buffers
614 */
615static void set_srp_direction(struct scsi_cmnd *cmd,
616			      struct srp_cmd *srp_cmd,
617			      int numbuf)
618{
619	u8 fmt;
620
621	if (numbuf == 0)
622		return;
623
624	if (numbuf == 1)
625		fmt = SRP_DATA_DESC_DIRECT;
626	else {
627		fmt = SRP_DATA_DESC_INDIRECT;
628		numbuf = min(numbuf, MAX_INDIRECT_BUFS);
629
630		if (cmd->sc_data_direction == DMA_TO_DEVICE)
631			srp_cmd->data_out_desc_cnt = numbuf;
632		else
633			srp_cmd->data_in_desc_cnt = numbuf;
634	}
635
636	if (cmd->sc_data_direction == DMA_TO_DEVICE)
637		srp_cmd->buf_fmt = fmt << 4;
638	else
639		srp_cmd->buf_fmt = fmt;
640}
641
642/**
643 * unmap_cmd_data: - Unmap data pointed in srp_cmd based on the format
644 * @cmd:	srp_cmd whose additional_data member will be unmapped
645 * @dev:	device for which the memory is mapped
646 *
647*/
648static void unmap_cmd_data(struct srp_cmd *cmd,
649			   struct srp_event_struct *evt_struct,
650			   struct device *dev)
651{
652	u8 out_fmt, in_fmt;
653
654	out_fmt = cmd->buf_fmt >> 4;
655	in_fmt = cmd->buf_fmt & ((1U << 4) - 1);
656
657	if (out_fmt == SRP_NO_DATA_DESC && in_fmt == SRP_NO_DATA_DESC)
658		return;
659
660	if (evt_struct->cmnd)
661		scsi_dma_unmap(evt_struct->cmnd);
662}
663
664static int map_sg_list(struct scsi_cmnd *cmd, int nseg,
665		       struct srp_direct_buf *md)
666{
667	int i;
668	struct scatterlist *sg;
669	u64 total_length = 0;
670
671	scsi_for_each_sg(cmd, sg, nseg, i) {
672		struct srp_direct_buf *descr = md + i;
673		descr->va = cpu_to_be64(sg_dma_address(sg));
674		descr->len = cpu_to_be32(sg_dma_len(sg));
675		descr->key = 0;
676		total_length += sg_dma_len(sg);
677 	}
678	return total_length;
679}
680
681/**
682 * map_sg_data: - Maps dma for a scatterlist and initializes decriptor fields
683 * @cmd:	Scsi_Cmnd with the scatterlist
684 * @srp_cmd:	srp_cmd that contains the memory descriptor
685 * @dev:	device for which to map dma memory
686 *
687 * Called by map_data_for_srp_cmd() when building srp cmd from scsi cmd.
688 * Returns 1 on success.
689*/
690static int map_sg_data(struct scsi_cmnd *cmd,
691		       struct srp_event_struct *evt_struct,
692		       struct srp_cmd *srp_cmd, struct device *dev)
693{
694
695	int sg_mapped;
696	u64 total_length = 0;
697	struct srp_direct_buf *data =
698		(struct srp_direct_buf *) srp_cmd->add_data;
699	struct srp_indirect_buf *indirect =
700		(struct srp_indirect_buf *) data;
701
702	sg_mapped = scsi_dma_map(cmd);
703	if (!sg_mapped)
704		return 1;
705	else if (sg_mapped < 0)
706		return 0;
707
708	set_srp_direction(cmd, srp_cmd, sg_mapped);
709
710	/* special case; we can use a single direct descriptor */
711	if (sg_mapped == 1) {
712		map_sg_list(cmd, sg_mapped, data);
713		return 1;
714	}
715
716	indirect->table_desc.va = 0;
717	indirect->table_desc.len = cpu_to_be32(sg_mapped *
718					       sizeof(struct srp_direct_buf));
719	indirect->table_desc.key = 0;
720
721	if (sg_mapped <= MAX_INDIRECT_BUFS) {
722		total_length = map_sg_list(cmd, sg_mapped,
723					   &indirect->desc_list[0]);
724		indirect->len = cpu_to_be32(total_length);
725		return 1;
726	}
727
728	/* get indirect table */
729	if (!evt_struct->ext_list) {
730		evt_struct->ext_list = (struct srp_direct_buf *)
731			dma_alloc_coherent(dev,
732					   SG_ALL * sizeof(struct srp_direct_buf),
733					   &evt_struct->ext_list_token, 0);
734		if (!evt_struct->ext_list) {
735			if (!firmware_has_feature(FW_FEATURE_CMO))
736				sdev_printk(KERN_ERR, cmd->device,
737				            "Can't allocate memory "
738				            "for indirect table\n");
739			scsi_dma_unmap(cmd);
740			return 0;
741		}
742	}
743
744	total_length = map_sg_list(cmd, sg_mapped, evt_struct->ext_list);
745
746	indirect->len = cpu_to_be32(total_length);
747	indirect->table_desc.va = cpu_to_be64(evt_struct->ext_list_token);
748	indirect->table_desc.len = cpu_to_be32(sg_mapped *
749					       sizeof(indirect->desc_list[0]));
750	memcpy(indirect->desc_list, evt_struct->ext_list,
751	       MAX_INDIRECT_BUFS * sizeof(struct srp_direct_buf));
752 	return 1;
753}
754
755/**
756 * map_data_for_srp_cmd: - Calls functions to map data for srp cmds
757 * @cmd:	struct scsi_cmnd with the memory to be mapped
758 * @srp_cmd:	srp_cmd that contains the memory descriptor
759 * @dev:	dma device for which to map dma memory
760 *
761 * Called by scsi_cmd_to_srp_cmd() when converting scsi cmds to srp cmds
762 * Returns 1 on success.
763*/
764static int map_data_for_srp_cmd(struct scsi_cmnd *cmd,
765				struct srp_event_struct *evt_struct,
766				struct srp_cmd *srp_cmd, struct device *dev)
767{
768	switch (cmd->sc_data_direction) {
769	case DMA_FROM_DEVICE:
770	case DMA_TO_DEVICE:
771		break;
772	case DMA_NONE:
773		return 1;
774	case DMA_BIDIRECTIONAL:
775		sdev_printk(KERN_ERR, cmd->device,
776			    "Can't map DMA_BIDIRECTIONAL to read/write\n");
777		return 0;
778	default:
779		sdev_printk(KERN_ERR, cmd->device,
780			    "Unknown data direction 0x%02x; can't map!\n",
781			    cmd->sc_data_direction);
782		return 0;
783	}
784
785	return map_sg_data(cmd, evt_struct, srp_cmd, dev);
786}
787
788/**
789 * purge_requests: Our virtual adapter just shut down.  purge any sent requests
790 * @hostdata:    the adapter
791 */
792static void purge_requests(struct ibmvscsi_host_data *hostdata, int error_code)
793{
794	struct srp_event_struct *evt;
795	unsigned long flags;
796
797	spin_lock_irqsave(hostdata->host->host_lock, flags);
798	while (!list_empty(&hostdata->sent)) {
799		evt = list_first_entry(&hostdata->sent, struct srp_event_struct, list);
800		list_del(&evt->list);
801		del_timer(&evt->timer);
802
803		spin_unlock_irqrestore(hostdata->host->host_lock, flags);
804		if (evt->cmnd) {
805			evt->cmnd->result = (error_code << 16);
806			unmap_cmd_data(&evt->iu.srp.cmd, evt,
807				       evt->hostdata->dev);
808			if (evt->cmnd_done)
809				evt->cmnd_done(evt->cmnd);
810		} else if (evt->done && evt->crq.format != VIOSRP_MAD_FORMAT &&
811			   evt->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
812			evt->done(evt);
813		free_event_struct(&evt->hostdata->pool, evt);
814		spin_lock_irqsave(hostdata->host->host_lock, flags);
815	}
816	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
817}
818
819/**
820 * ibmvscsi_reset_host - Reset the connection to the server
821 * @hostdata:	struct ibmvscsi_host_data to reset
822*/
823static void ibmvscsi_reset_host(struct ibmvscsi_host_data *hostdata)
824{
825	scsi_block_requests(hostdata->host);
826	atomic_set(&hostdata->request_limit, 0);
827
828	purge_requests(hostdata, DID_ERROR);
829	hostdata->reset_crq = 1;
830	wake_up(&hostdata->work_wait_q);
831}
832
833/**
834 * ibmvscsi_timeout - Internal command timeout handler
835 * @evt_struct:	struct srp_event_struct that timed out
836 *
837 * Called when an internally generated command times out
838*/
839static void ibmvscsi_timeout(struct srp_event_struct *evt_struct)
840{
841	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
842
843	dev_err(hostdata->dev, "Command timed out (%x). Resetting connection\n",
844		evt_struct->iu.srp.cmd.opcode);
845
846	ibmvscsi_reset_host(hostdata);
847}
848
849
850/* ------------------------------------------------------------
851 * Routines for sending and receiving SRPs
852 */
853/**
854 * ibmvscsi_send_srp_event: - Transforms event to u64 array and calls send_crq()
855 * @evt_struct:	evt_struct to be sent
856 * @hostdata:	ibmvscsi_host_data of host
857 * @timeout:	timeout in seconds - 0 means do not time command
858 *
859 * Returns the value returned from ibmvscsi_send_crq(). (Zero for success)
860 * Note that this routine assumes that host_lock is held for synchronization
861*/
862static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
863				   struct ibmvscsi_host_data *hostdata,
864				   unsigned long timeout)
865{
866	__be64 *crq_as_u64 = (__be64 *)&evt_struct->crq;
867	int request_status = 0;
868	int rc;
869	int srp_req = 0;
870
871	/* If we have exhausted our request limit, just fail this request,
872	 * unless it is for a reset or abort.
873	 * Note that there are rare cases involving driver generated requests
874	 * (such as task management requests) that the mid layer may think we
875	 * can handle more requests (can_queue) when we actually can't
876	 */
877	if (evt_struct->crq.format == VIOSRP_SRP_FORMAT) {
878		srp_req = 1;
879		request_status =
880			atomic_dec_if_positive(&hostdata->request_limit);
881		/* If request limit was -1 when we started, it is now even
882		 * less than that
883		 */
884		if (request_status < -1)
885			goto send_error;
886		/* Otherwise, we may have run out of requests. */
887		/* If request limit was 0 when we started the adapter is in the
888		 * process of performing a login with the server adapter, or
889		 * we may have run out of requests.
890		 */
891		else if (request_status == -1 &&
892		         evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
893			goto send_busy;
894		/* Abort and reset calls should make it through.
895		 * Nothing except abort and reset should use the last two
896		 * slots unless we had two or less to begin with.
897		 */
898		else if (request_status < 2 &&
899		         evt_struct->iu.srp.cmd.opcode != SRP_TSK_MGMT) {
900			/* In the case that we have less than two requests
901			 * available, check the server limit as a combination
902			 * of the request limit and the number of requests
903			 * in-flight (the size of the send list).  If the
904			 * server limit is greater than 2, return busy so
905			 * that the last two are reserved for reset and abort.
906			 */
907			int server_limit = request_status;
908			struct srp_event_struct *tmp_evt;
909
910			list_for_each_entry(tmp_evt, &hostdata->sent, list) {
911				server_limit++;
912			}
913
914			if (server_limit > 2)
915				goto send_busy;
916		}
917	}
918
919	/* Copy the IU into the transfer area */
920	*evt_struct->xfer_iu = evt_struct->iu;
921	evt_struct->xfer_iu->srp.rsp.tag = (u64)evt_struct;
922
923	/* Add this to the sent list.  We need to do this
924	 * before we actually send
925	 * in case it comes back REALLY fast
926	 */
927	list_add_tail(&evt_struct->list, &hostdata->sent);
928
929	init_timer(&evt_struct->timer);
930	if (timeout) {
931		evt_struct->timer.data = (unsigned long) evt_struct;
932		evt_struct->timer.expires = jiffies + (timeout * HZ);
933		evt_struct->timer.function = (void (*)(unsigned long))ibmvscsi_timeout;
934		add_timer(&evt_struct->timer);
935	}
936
937	rc = ibmvscsi_send_crq(hostdata, be64_to_cpu(crq_as_u64[0]),
938			       be64_to_cpu(crq_as_u64[1]));
939	if (rc != 0) {
940		list_del(&evt_struct->list);
941		del_timer(&evt_struct->timer);
942
943		/* If send_crq returns H_CLOSED, return SCSI_MLQUEUE_HOST_BUSY.
944		 * Firmware will send a CRQ with a transport event (0xFF) to
945		 * tell this client what has happened to the transport.  This
946		 * will be handled in ibmvscsi_handle_crq()
947		 */
948		if (rc == H_CLOSED) {
949			dev_warn(hostdata->dev, "send warning. "
950			         "Receive queue closed, will retry.\n");
951			goto send_busy;
952		}
953		dev_err(hostdata->dev, "send error %d\n", rc);
954		if (srp_req)
955			atomic_inc(&hostdata->request_limit);
956		goto send_error;
957	}
958
959	return 0;
960
961 send_busy:
962	unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
963
964	free_event_struct(&hostdata->pool, evt_struct);
965	if (srp_req && request_status != -1)
966		atomic_inc(&hostdata->request_limit);
967	return SCSI_MLQUEUE_HOST_BUSY;
968
969 send_error:
970	unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
971
972	if (evt_struct->cmnd != NULL) {
973		evt_struct->cmnd->result = DID_ERROR << 16;
974		evt_struct->cmnd_done(evt_struct->cmnd);
975	} else if (evt_struct->done)
976		evt_struct->done(evt_struct);
977
978	free_event_struct(&hostdata->pool, evt_struct);
979	return 0;
980}
981
982/**
983 * handle_cmd_rsp: -  Handle responses from commands
984 * @evt_struct:	srp_event_struct to be handled
985 *
986 * Used as a callback by when sending scsi cmds.
987 * Gets called by ibmvscsi_handle_crq()
988*/
989static void handle_cmd_rsp(struct srp_event_struct *evt_struct)
990{
991	struct srp_rsp *rsp = &evt_struct->xfer_iu->srp.rsp;
992	struct scsi_cmnd *cmnd = evt_struct->cmnd;
993
994	if (unlikely(rsp->opcode != SRP_RSP)) {
995		if (printk_ratelimit())
996			dev_warn(evt_struct->hostdata->dev,
997				 "bad SRP RSP type %d\n", rsp->opcode);
998	}
999
1000	if (cmnd) {
1001		cmnd->result |= rsp->status;
1002		if (((cmnd->result >> 1) & 0x1f) == CHECK_CONDITION)
1003			memcpy(cmnd->sense_buffer,
1004			       rsp->data,
1005			       be32_to_cpu(rsp->sense_data_len));
1006		unmap_cmd_data(&evt_struct->iu.srp.cmd,
1007			       evt_struct,
1008			       evt_struct->hostdata->dev);
1009
1010		if (rsp->flags & SRP_RSP_FLAG_DOOVER)
1011			scsi_set_resid(cmnd,
1012				       be32_to_cpu(rsp->data_out_res_cnt));
1013		else if (rsp->flags & SRP_RSP_FLAG_DIOVER)
1014			scsi_set_resid(cmnd, be32_to_cpu(rsp->data_in_res_cnt));
1015	}
1016
1017	if (evt_struct->cmnd_done)
1018		evt_struct->cmnd_done(cmnd);
1019}
1020
1021/**
1022 * lun_from_dev: - Returns the lun of the scsi device
1023 * @dev:	struct scsi_device
1024 *
1025*/
1026static inline u16 lun_from_dev(struct scsi_device *dev)
1027{
1028	return (0x2 << 14) | (dev->id << 8) | (dev->channel << 5) | dev->lun;
1029}
1030
1031/**
1032 * ibmvscsi_queue: - The queuecommand function of the scsi template
1033 * @cmd:	struct scsi_cmnd to be executed
1034 * @done:	Callback function to be called when cmd is completed
1035*/
1036static int ibmvscsi_queuecommand_lck(struct scsi_cmnd *cmnd,
1037				 void (*done) (struct scsi_cmnd *))
1038{
1039	struct srp_cmd *srp_cmd;
1040	struct srp_event_struct *evt_struct;
1041	struct srp_indirect_buf *indirect;
1042	struct ibmvscsi_host_data *hostdata = shost_priv(cmnd->device->host);
1043	u16 lun = lun_from_dev(cmnd->device);
1044	u8 out_fmt, in_fmt;
1045
1046	cmnd->result = (DID_OK << 16);
1047	evt_struct = get_event_struct(&hostdata->pool);
1048	if (!evt_struct)
1049		return SCSI_MLQUEUE_HOST_BUSY;
1050
1051	/* Set up the actual SRP IU */
1052	srp_cmd = &evt_struct->iu.srp.cmd;
1053	memset(srp_cmd, 0x00, SRP_MAX_IU_LEN);
1054	srp_cmd->opcode = SRP_CMD;
1055	memcpy(srp_cmd->cdb, cmnd->cmnd, sizeof(srp_cmd->cdb));
1056	srp_cmd->lun = cpu_to_be64(((u64)lun) << 48);
1057
1058	if (!map_data_for_srp_cmd(cmnd, evt_struct, srp_cmd, hostdata->dev)) {
1059		if (!firmware_has_feature(FW_FEATURE_CMO))
1060			sdev_printk(KERN_ERR, cmnd->device,
1061			            "couldn't convert cmd to srp_cmd\n");
1062		free_event_struct(&hostdata->pool, evt_struct);
1063		return SCSI_MLQUEUE_HOST_BUSY;
1064	}
1065
1066	init_event_struct(evt_struct,
1067			  handle_cmd_rsp,
1068			  VIOSRP_SRP_FORMAT,
1069			  cmnd->request->timeout/HZ);
1070
1071	evt_struct->cmnd = cmnd;
1072	evt_struct->cmnd_done = done;
1073
1074	/* Fix up dma address of the buffer itself */
1075	indirect = (struct srp_indirect_buf *) srp_cmd->add_data;
1076	out_fmt = srp_cmd->buf_fmt >> 4;
1077	in_fmt = srp_cmd->buf_fmt & ((1U << 4) - 1);
1078	if ((in_fmt == SRP_DATA_DESC_INDIRECT ||
1079	     out_fmt == SRP_DATA_DESC_INDIRECT) &&
1080	    indirect->table_desc.va == 0) {
1081		indirect->table_desc.va =
1082			cpu_to_be64(be64_to_cpu(evt_struct->crq.IU_data_ptr) +
1083			offsetof(struct srp_cmd, add_data) +
1084			offsetof(struct srp_indirect_buf, desc_list));
1085	}
1086
1087	return ibmvscsi_send_srp_event(evt_struct, hostdata, 0);
1088}
1089
1090static DEF_SCSI_QCMD(ibmvscsi_queuecommand)
1091
1092/* ------------------------------------------------------------
1093 * Routines for driver initialization
1094 */
1095
1096/**
1097 * map_persist_bufs: - Pre-map persistent data for adapter logins
1098 * @hostdata:   ibmvscsi_host_data of host
1099 *
1100 * Map the capabilities and adapter info DMA buffers to avoid runtime failures.
1101 * Return 1 on error, 0 on success.
1102 */
1103static int map_persist_bufs(struct ibmvscsi_host_data *hostdata)
1104{
1105
1106	hostdata->caps_addr = dma_map_single(hostdata->dev, &hostdata->caps,
1107					     sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1108
1109	if (dma_mapping_error(hostdata->dev, hostdata->caps_addr)) {
1110		dev_err(hostdata->dev, "Unable to map capabilities buffer!\n");
1111		return 1;
1112	}
1113
1114	hostdata->adapter_info_addr = dma_map_single(hostdata->dev,
1115						     &hostdata->madapter_info,
1116						     sizeof(hostdata->madapter_info),
1117						     DMA_BIDIRECTIONAL);
1118	if (dma_mapping_error(hostdata->dev, hostdata->adapter_info_addr)) {
1119		dev_err(hostdata->dev, "Unable to map adapter info buffer!\n");
1120		dma_unmap_single(hostdata->dev, hostdata->caps_addr,
1121				 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1122		return 1;
1123	}
1124
1125	return 0;
1126}
1127
1128/**
1129 * unmap_persist_bufs: - Unmap persistent data needed for adapter logins
1130 * @hostdata:   ibmvscsi_host_data of host
1131 *
1132 * Unmap the capabilities and adapter info DMA buffers
1133 */
1134static void unmap_persist_bufs(struct ibmvscsi_host_data *hostdata)
1135{
1136	dma_unmap_single(hostdata->dev, hostdata->caps_addr,
1137			 sizeof(hostdata->caps), DMA_BIDIRECTIONAL);
1138
1139	dma_unmap_single(hostdata->dev, hostdata->adapter_info_addr,
1140			 sizeof(hostdata->madapter_info), DMA_BIDIRECTIONAL);
1141}
1142
1143/**
1144 * login_rsp: - Handle response to SRP login request
1145 * @evt_struct:	srp_event_struct with the response
1146 *
1147 * Used as a "done" callback by when sending srp_login. Gets called
1148 * by ibmvscsi_handle_crq()
1149*/
1150static void login_rsp(struct srp_event_struct *evt_struct)
1151{
1152	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1153	switch (evt_struct->xfer_iu->srp.login_rsp.opcode) {
1154	case SRP_LOGIN_RSP:	/* it worked! */
1155		break;
1156	case SRP_LOGIN_REJ:	/* refused! */
1157		dev_info(hostdata->dev, "SRP_LOGIN_REJ reason %u\n",
1158			 evt_struct->xfer_iu->srp.login_rej.reason);
1159		/* Login failed.  */
1160		atomic_set(&hostdata->request_limit, -1);
1161		return;
1162	default:
1163		dev_err(hostdata->dev, "Invalid login response typecode 0x%02x!\n",
1164			evt_struct->xfer_iu->srp.login_rsp.opcode);
1165		/* Login failed.  */
1166		atomic_set(&hostdata->request_limit, -1);
1167		return;
1168	}
1169
1170	dev_info(hostdata->dev, "SRP_LOGIN succeeded\n");
1171	hostdata->client_migrated = 0;
1172
1173	/* Now we know what the real request-limit is.
1174	 * This value is set rather than added to request_limit because
1175	 * request_limit could have been set to -1 by this client.
1176	 */
1177	atomic_set(&hostdata->request_limit,
1178		   be32_to_cpu(evt_struct->xfer_iu->srp.login_rsp.req_lim_delta));
1179
1180	/* If we had any pending I/Os, kick them */
1181	scsi_unblock_requests(hostdata->host);
1182}
1183
1184/**
1185 * send_srp_login: - Sends the srp login
1186 * @hostdata:	ibmvscsi_host_data of host
1187 *
1188 * Returns zero if successful.
1189*/
1190static int send_srp_login(struct ibmvscsi_host_data *hostdata)
1191{
1192	int rc;
1193	unsigned long flags;
1194	struct srp_login_req *login;
1195	struct srp_event_struct *evt_struct = get_event_struct(&hostdata->pool);
1196
1197	BUG_ON(!evt_struct);
1198	init_event_struct(evt_struct, login_rsp,
1199			  VIOSRP_SRP_FORMAT, login_timeout);
1200
1201	login = &evt_struct->iu.srp.login_req;
1202	memset(login, 0, sizeof(*login));
1203	login->opcode = SRP_LOGIN_REQ;
1204	login->req_it_iu_len = cpu_to_be32(sizeof(union srp_iu));
1205	login->req_buf_fmt = cpu_to_be16(SRP_BUF_FORMAT_DIRECT |
1206					 SRP_BUF_FORMAT_INDIRECT);
1207
1208	spin_lock_irqsave(hostdata->host->host_lock, flags);
1209	/* Start out with a request limit of 0, since this is negotiated in
1210	 * the login request we are just sending and login requests always
1211	 * get sent by the driver regardless of request_limit.
1212	 */
1213	atomic_set(&hostdata->request_limit, 0);
1214
1215	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, login_timeout * 2);
1216	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1217	dev_info(hostdata->dev, "sent SRP login\n");
1218	return rc;
1219};
1220
1221/**
1222 * capabilities_rsp: - Handle response to MAD adapter capabilities request
1223 * @evt_struct:	srp_event_struct with the response
1224 *
1225 * Used as a "done" callback by when sending adapter_info.
1226 */
1227static void capabilities_rsp(struct srp_event_struct *evt_struct)
1228{
1229	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1230
1231	if (evt_struct->xfer_iu->mad.capabilities.common.status) {
1232		dev_err(hostdata->dev, "error 0x%X getting capabilities info\n",
1233			evt_struct->xfer_iu->mad.capabilities.common.status);
1234	} else {
1235		if (hostdata->caps.migration.common.server_support !=
1236		    cpu_to_be16(SERVER_SUPPORTS_CAP))
1237			dev_info(hostdata->dev, "Partition migration not supported\n");
1238
1239		if (client_reserve) {
1240			if (hostdata->caps.reserve.common.server_support ==
1241			    cpu_to_be16(SERVER_SUPPORTS_CAP))
1242				dev_info(hostdata->dev, "Client reserve enabled\n");
1243			else
1244				dev_info(hostdata->dev, "Client reserve not supported\n");
1245		}
1246	}
1247
1248	send_srp_login(hostdata);
1249}
1250
1251/**
1252 * send_mad_capabilities: - Sends the mad capabilities request
1253 *      and stores the result so it can be retrieved with
1254 * @hostdata:	ibmvscsi_host_data of host
1255 */
1256static void send_mad_capabilities(struct ibmvscsi_host_data *hostdata)
1257{
1258	struct viosrp_capabilities *req;
1259	struct srp_event_struct *evt_struct;
1260	unsigned long flags;
1261	struct device_node *of_node = hostdata->dev->of_node;
1262	const char *location;
1263
1264	evt_struct = get_event_struct(&hostdata->pool);
1265	BUG_ON(!evt_struct);
1266
1267	init_event_struct(evt_struct, capabilities_rsp,
1268			  VIOSRP_MAD_FORMAT, info_timeout);
1269
1270	req = &evt_struct->iu.mad.capabilities;
1271	memset(req, 0, sizeof(*req));
1272
1273	hostdata->caps.flags = cpu_to_be32(CAP_LIST_SUPPORTED);
1274	if (hostdata->client_migrated)
1275		hostdata->caps.flags |= cpu_to_be32(CLIENT_MIGRATED);
1276
1277	strncpy(hostdata->caps.name, dev_name(&hostdata->host->shost_gendev),
1278		sizeof(hostdata->caps.name));
1279	hostdata->caps.name[sizeof(hostdata->caps.name) - 1] = '\0';
1280
1281	location = of_get_property(of_node, "ibm,loc-code", NULL);
1282	location = location ? location : dev_name(hostdata->dev);
1283	strncpy(hostdata->caps.loc, location, sizeof(hostdata->caps.loc));
1284	hostdata->caps.loc[sizeof(hostdata->caps.loc) - 1] = '\0';
1285
1286	req->common.type = cpu_to_be32(VIOSRP_CAPABILITIES_TYPE);
1287	req->buffer = cpu_to_be64(hostdata->caps_addr);
1288
1289	hostdata->caps.migration.common.cap_type =
1290				cpu_to_be32(MIGRATION_CAPABILITIES);
1291	hostdata->caps.migration.common.length =
1292				cpu_to_be16(sizeof(hostdata->caps.migration));
1293	hostdata->caps.migration.common.server_support =
1294				cpu_to_be16(SERVER_SUPPORTS_CAP);
1295	hostdata->caps.migration.ecl = cpu_to_be32(1);
1296
1297	if (client_reserve) {
1298		hostdata->caps.reserve.common.cap_type =
1299					cpu_to_be32(RESERVATION_CAPABILITIES);
1300		hostdata->caps.reserve.common.length =
1301				cpu_to_be16(sizeof(hostdata->caps.reserve));
1302		hostdata->caps.reserve.common.server_support =
1303				cpu_to_be16(SERVER_SUPPORTS_CAP);
1304		hostdata->caps.reserve.type =
1305				cpu_to_be32(CLIENT_RESERVE_SCSI_2);
1306		req->common.length =
1307				cpu_to_be16(sizeof(hostdata->caps));
1308	} else
1309		req->common.length = cpu_to_be16(sizeof(hostdata->caps) -
1310						sizeof(hostdata->caps.reserve));
1311
1312	spin_lock_irqsave(hostdata->host->host_lock, flags);
1313	if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1314		dev_err(hostdata->dev, "couldn't send CAPABILITIES_REQ!\n");
1315	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1316};
1317
1318/**
1319 * fast_fail_rsp: - Handle response to MAD enable fast fail
1320 * @evt_struct:	srp_event_struct with the response
1321 *
1322 * Used as a "done" callback by when sending enable fast fail. Gets called
1323 * by ibmvscsi_handle_crq()
1324 */
1325static void fast_fail_rsp(struct srp_event_struct *evt_struct)
1326{
1327	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1328	u16 status = be16_to_cpu(evt_struct->xfer_iu->mad.fast_fail.common.status);
1329
1330	if (status == VIOSRP_MAD_NOT_SUPPORTED)
1331		dev_err(hostdata->dev, "fast_fail not supported in server\n");
1332	else if (status == VIOSRP_MAD_FAILED)
1333		dev_err(hostdata->dev, "fast_fail request failed\n");
1334	else if (status != VIOSRP_MAD_SUCCESS)
1335		dev_err(hostdata->dev, "error 0x%X enabling fast_fail\n", status);
1336
1337	send_mad_capabilities(hostdata);
1338}
1339
1340/**
1341 * init_host - Start host initialization
1342 * @hostdata:	ibmvscsi_host_data of host
1343 *
1344 * Returns zero if successful.
1345 */
1346static int enable_fast_fail(struct ibmvscsi_host_data *hostdata)
1347{
1348	int rc;
1349	unsigned long flags;
1350	struct viosrp_fast_fail *fast_fail_mad;
1351	struct srp_event_struct *evt_struct;
1352
1353	if (!fast_fail) {
1354		send_mad_capabilities(hostdata);
1355		return 0;
1356	}
1357
1358	evt_struct = get_event_struct(&hostdata->pool);
1359	BUG_ON(!evt_struct);
1360
1361	init_event_struct(evt_struct, fast_fail_rsp, VIOSRP_MAD_FORMAT, info_timeout);
1362
1363	fast_fail_mad = &evt_struct->iu.mad.fast_fail;
1364	memset(fast_fail_mad, 0, sizeof(*fast_fail_mad));
1365	fast_fail_mad->common.type = cpu_to_be32(VIOSRP_ENABLE_FAST_FAIL);
1366	fast_fail_mad->common.length = cpu_to_be16(sizeof(*fast_fail_mad));
1367
1368	spin_lock_irqsave(hostdata->host->host_lock, flags);
1369	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1370	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1371	return rc;
1372}
1373
1374/**
1375 * adapter_info_rsp: - Handle response to MAD adapter info request
1376 * @evt_struct:	srp_event_struct with the response
1377 *
1378 * Used as a "done" callback by when sending adapter_info. Gets called
1379 * by ibmvscsi_handle_crq()
1380*/
1381static void adapter_info_rsp(struct srp_event_struct *evt_struct)
1382{
1383	struct ibmvscsi_host_data *hostdata = evt_struct->hostdata;
1384
1385	if (evt_struct->xfer_iu->mad.adapter_info.common.status) {
1386		dev_err(hostdata->dev, "error %d getting adapter info\n",
1387			evt_struct->xfer_iu->mad.adapter_info.common.status);
1388	} else {
1389		dev_info(hostdata->dev, "host srp version: %s, "
1390			 "host partition %s (%d), OS %d, max io %u\n",
1391			 hostdata->madapter_info.srp_version,
1392			 hostdata->madapter_info.partition_name,
1393			 be32_to_cpu(hostdata->madapter_info.partition_number),
1394			 be32_to_cpu(hostdata->madapter_info.os_type),
1395			 be32_to_cpu(hostdata->madapter_info.port_max_txu[0]));
1396
1397		if (hostdata->madapter_info.port_max_txu[0])
1398			hostdata->host->max_sectors =
1399				be32_to_cpu(hostdata->madapter_info.port_max_txu[0]) >> 9;
1400
1401		if (be32_to_cpu(hostdata->madapter_info.os_type) == 3 &&
1402		    strcmp(hostdata->madapter_info.srp_version, "1.6a") <= 0) {
1403			dev_err(hostdata->dev, "host (Ver. %s) doesn't support large transfers\n",
1404				hostdata->madapter_info.srp_version);
1405			dev_err(hostdata->dev, "limiting scatterlists to %d\n",
1406				MAX_INDIRECT_BUFS);
1407			hostdata->host->sg_tablesize = MAX_INDIRECT_BUFS;
1408		}
1409
1410		if (be32_to_cpu(hostdata->madapter_info.os_type) == 3) {
1411			enable_fast_fail(hostdata);
1412			return;
1413		}
1414	}
1415
1416	send_srp_login(hostdata);
1417}
1418
1419/**
1420 * send_mad_adapter_info: - Sends the mad adapter info request
1421 *      and stores the result so it can be retrieved with
1422 *      sysfs.  We COULD consider causing a failure if the
1423 *      returned SRP version doesn't match ours.
1424 * @hostdata:	ibmvscsi_host_data of host
1425 *
1426 * Returns zero if successful.
1427*/
1428static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
1429{
1430	struct viosrp_adapter_info *req;
1431	struct srp_event_struct *evt_struct;
1432	unsigned long flags;
1433
1434	evt_struct = get_event_struct(&hostdata->pool);
1435	BUG_ON(!evt_struct);
1436
1437	init_event_struct(evt_struct,
1438			  adapter_info_rsp,
1439			  VIOSRP_MAD_FORMAT,
1440			  info_timeout);
1441
1442	req = &evt_struct->iu.mad.adapter_info;
1443	memset(req, 0x00, sizeof(*req));
1444
1445	req->common.type = cpu_to_be32(VIOSRP_ADAPTER_INFO_TYPE);
1446	req->common.length = cpu_to_be16(sizeof(hostdata->madapter_info));
1447	req->buffer = cpu_to_be64(hostdata->adapter_info_addr);
1448
1449	spin_lock_irqsave(hostdata->host->host_lock, flags);
1450	if (ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2))
1451		dev_err(hostdata->dev, "couldn't send ADAPTER_INFO_REQ!\n");
1452	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1453};
1454
1455/**
1456 * init_adapter: Start virtual adapter initialization sequence
1457 *
1458 */
1459static void init_adapter(struct ibmvscsi_host_data *hostdata)
1460{
1461	send_mad_adapter_info(hostdata);
1462}
1463
1464/**
1465 * sync_completion: Signal that a synchronous command has completed
1466 * Note that after returning from this call, the evt_struct is freed.
1467 * the caller waiting on this completion shouldn't touch the evt_struct
1468 * again.
1469 */
1470static void sync_completion(struct srp_event_struct *evt_struct)
1471{
1472	/* copy the response back */
1473	if (evt_struct->sync_srp)
1474		*evt_struct->sync_srp = *evt_struct->xfer_iu;
1475
1476	complete(&evt_struct->comp);
1477}
1478
1479/**
1480 * ibmvscsi_abort: Abort a command...from scsi host template
1481 * send this over to the server and wait synchronously for the response
1482 */
1483static int ibmvscsi_eh_abort_handler(struct scsi_cmnd *cmd)
1484{
1485	struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1486	struct srp_tsk_mgmt *tsk_mgmt;
1487	struct srp_event_struct *evt;
1488	struct srp_event_struct *tmp_evt, *found_evt;
1489	union viosrp_iu srp_rsp;
1490	int rsp_rc;
1491	unsigned long flags;
1492	u16 lun = lun_from_dev(cmd->device);
1493	unsigned long wait_switch = 0;
1494
1495	/* First, find this command in our sent list so we can figure
1496	 * out the correct tag
1497	 */
1498	spin_lock_irqsave(hostdata->host->host_lock, flags);
1499	wait_switch = jiffies + (init_timeout * HZ);
1500	do {
1501		found_evt = NULL;
1502		list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1503			if (tmp_evt->cmnd == cmd) {
1504				found_evt = tmp_evt;
1505				break;
1506			}
1507		}
1508
1509		if (!found_evt) {
1510			spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1511			return SUCCESS;
1512		}
1513
1514		evt = get_event_struct(&hostdata->pool);
1515		if (evt == NULL) {
1516			spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1517			sdev_printk(KERN_ERR, cmd->device,
1518				"failed to allocate abort event\n");
1519			return FAILED;
1520		}
1521
1522		init_event_struct(evt,
1523				  sync_completion,
1524				  VIOSRP_SRP_FORMAT,
1525				  abort_timeout);
1526
1527		tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1528
1529		/* Set up an abort SRP command */
1530		memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1531		tsk_mgmt->opcode = SRP_TSK_MGMT;
1532		tsk_mgmt->lun = cpu_to_be64(((u64) lun) << 48);
1533		tsk_mgmt->tsk_mgmt_func = SRP_TSK_ABORT_TASK;
1534		tsk_mgmt->task_tag = (u64) found_evt;
1535
1536		evt->sync_srp = &srp_rsp;
1537
1538		init_completion(&evt->comp);
1539		rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, abort_timeout * 2);
1540
1541		if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1542			break;
1543
1544		spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1545		msleep(10);
1546		spin_lock_irqsave(hostdata->host->host_lock, flags);
1547	} while (time_before(jiffies, wait_switch));
1548
1549	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1550
1551	if (rsp_rc != 0) {
1552		sdev_printk(KERN_ERR, cmd->device,
1553			    "failed to send abort() event. rc=%d\n", rsp_rc);
1554		return FAILED;
1555	}
1556
1557	sdev_printk(KERN_INFO, cmd->device,
1558                    "aborting command. lun 0x%llx, tag 0x%llx\n",
1559		    (((u64) lun) << 48), (u64) found_evt);
1560
1561	wait_for_completion(&evt->comp);
1562
1563	/* make sure we got a good response */
1564	if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1565		if (printk_ratelimit())
1566			sdev_printk(KERN_WARNING, cmd->device, "abort bad SRP RSP type %d\n",
1567				    srp_rsp.srp.rsp.opcode);
1568		return FAILED;
1569	}
1570
1571	if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1572		rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1573	else
1574		rsp_rc = srp_rsp.srp.rsp.status;
1575
1576	if (rsp_rc) {
1577		if (printk_ratelimit())
1578			sdev_printk(KERN_WARNING, cmd->device,
1579				    "abort code %d for task tag 0x%llx\n",
1580				    rsp_rc, tsk_mgmt->task_tag);
1581		return FAILED;
1582	}
1583
1584	/* Because we dropped the spinlock above, it's possible
1585	 * The event is no longer in our list.  Make sure it didn't
1586	 * complete while we were aborting
1587	 */
1588	spin_lock_irqsave(hostdata->host->host_lock, flags);
1589	found_evt = NULL;
1590	list_for_each_entry(tmp_evt, &hostdata->sent, list) {
1591		if (tmp_evt->cmnd == cmd) {
1592			found_evt = tmp_evt;
1593			break;
1594		}
1595	}
1596
1597	if (found_evt == NULL) {
1598		spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1599		sdev_printk(KERN_INFO, cmd->device, "aborted task tag 0x%llx completed\n",
1600			    tsk_mgmt->task_tag);
1601		return SUCCESS;
1602	}
1603
1604	sdev_printk(KERN_INFO, cmd->device, "successfully aborted task tag 0x%llx\n",
1605		    tsk_mgmt->task_tag);
1606
1607	cmd->result = (DID_ABORT << 16);
1608	list_del(&found_evt->list);
1609	unmap_cmd_data(&found_evt->iu.srp.cmd, found_evt,
1610		       found_evt->hostdata->dev);
1611	free_event_struct(&found_evt->hostdata->pool, found_evt);
1612	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1613	atomic_inc(&hostdata->request_limit);
1614	return SUCCESS;
1615}
1616
1617/**
1618 * ibmvscsi_eh_device_reset_handler: Reset a single LUN...from scsi host
1619 * template send this over to the server and wait synchronously for the
1620 * response
1621 */
1622static int ibmvscsi_eh_device_reset_handler(struct scsi_cmnd *cmd)
1623{
1624	struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1625	struct srp_tsk_mgmt *tsk_mgmt;
1626	struct srp_event_struct *evt;
1627	struct srp_event_struct *tmp_evt, *pos;
1628	union viosrp_iu srp_rsp;
1629	int rsp_rc;
1630	unsigned long flags;
1631	u16 lun = lun_from_dev(cmd->device);
1632	unsigned long wait_switch = 0;
1633
1634	spin_lock_irqsave(hostdata->host->host_lock, flags);
1635	wait_switch = jiffies + (init_timeout * HZ);
1636	do {
1637		evt = get_event_struct(&hostdata->pool);
1638		if (evt == NULL) {
1639			spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1640			sdev_printk(KERN_ERR, cmd->device,
1641				"failed to allocate reset event\n");
1642			return FAILED;
1643		}
1644
1645		init_event_struct(evt,
1646				  sync_completion,
1647				  VIOSRP_SRP_FORMAT,
1648				  reset_timeout);
1649
1650		tsk_mgmt = &evt->iu.srp.tsk_mgmt;
1651
1652		/* Set up a lun reset SRP command */
1653		memset(tsk_mgmt, 0x00, sizeof(*tsk_mgmt));
1654		tsk_mgmt->opcode = SRP_TSK_MGMT;
1655		tsk_mgmt->lun = cpu_to_be64(((u64) lun) << 48);
1656		tsk_mgmt->tsk_mgmt_func = SRP_TSK_LUN_RESET;
1657
1658		evt->sync_srp = &srp_rsp;
1659
1660		init_completion(&evt->comp);
1661		rsp_rc = ibmvscsi_send_srp_event(evt, hostdata, reset_timeout * 2);
1662
1663		if (rsp_rc != SCSI_MLQUEUE_HOST_BUSY)
1664			break;
1665
1666		spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1667		msleep(10);
1668		spin_lock_irqsave(hostdata->host->host_lock, flags);
1669	} while (time_before(jiffies, wait_switch));
1670
1671	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1672
1673	if (rsp_rc != 0) {
1674		sdev_printk(KERN_ERR, cmd->device,
1675			    "failed to send reset event. rc=%d\n", rsp_rc);
1676		return FAILED;
1677	}
1678
1679	sdev_printk(KERN_INFO, cmd->device, "resetting device. lun 0x%llx\n",
1680		    (((u64) lun) << 48));
1681
1682	wait_for_completion(&evt->comp);
1683
1684	/* make sure we got a good response */
1685	if (unlikely(srp_rsp.srp.rsp.opcode != SRP_RSP)) {
1686		if (printk_ratelimit())
1687			sdev_printk(KERN_WARNING, cmd->device, "reset bad SRP RSP type %d\n",
1688				    srp_rsp.srp.rsp.opcode);
1689		return FAILED;
1690	}
1691
1692	if (srp_rsp.srp.rsp.flags & SRP_RSP_FLAG_RSPVALID)
1693		rsp_rc = *((int *)srp_rsp.srp.rsp.data);
1694	else
1695		rsp_rc = srp_rsp.srp.rsp.status;
1696
1697	if (rsp_rc) {
1698		if (printk_ratelimit())
1699			sdev_printk(KERN_WARNING, cmd->device,
1700				    "reset code %d for task tag 0x%llx\n",
1701				    rsp_rc, tsk_mgmt->task_tag);
1702		return FAILED;
1703	}
1704
1705	/* We need to find all commands for this LUN that have not yet been
1706	 * responded to, and fail them with DID_RESET
1707	 */
1708	spin_lock_irqsave(hostdata->host->host_lock, flags);
1709	list_for_each_entry_safe(tmp_evt, pos, &hostdata->sent, list) {
1710		if ((tmp_evt->cmnd) && (tmp_evt->cmnd->device == cmd->device)) {
1711			if (tmp_evt->cmnd)
1712				tmp_evt->cmnd->result = (DID_RESET << 16);
1713			list_del(&tmp_evt->list);
1714			unmap_cmd_data(&tmp_evt->iu.srp.cmd, tmp_evt,
1715				       tmp_evt->hostdata->dev);
1716			free_event_struct(&tmp_evt->hostdata->pool,
1717						   tmp_evt);
1718			atomic_inc(&hostdata->request_limit);
1719			if (tmp_evt->cmnd_done)
1720				tmp_evt->cmnd_done(tmp_evt->cmnd);
1721			else if (tmp_evt->done)
1722				tmp_evt->done(tmp_evt);
1723		}
1724	}
1725	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1726	return SUCCESS;
1727}
1728
1729/**
1730 * ibmvscsi_eh_host_reset_handler - Reset the connection to the server
1731 * @cmd:	struct scsi_cmnd having problems
1732*/
1733static int ibmvscsi_eh_host_reset_handler(struct scsi_cmnd *cmd)
1734{
1735	unsigned long wait_switch = 0;
1736	struct ibmvscsi_host_data *hostdata = shost_priv(cmd->device->host);
1737
1738	dev_err(hostdata->dev, "Resetting connection due to error recovery\n");
1739
1740	ibmvscsi_reset_host(hostdata);
1741
1742	for (wait_switch = jiffies + (init_timeout * HZ);
1743	     time_before(jiffies, wait_switch) &&
1744		     atomic_read(&hostdata->request_limit) < 2;) {
1745
1746		msleep(10);
1747	}
1748
1749	if (atomic_read(&hostdata->request_limit) <= 0)
1750		return FAILED;
1751
1752	return SUCCESS;
1753}
1754
1755/**
1756 * ibmvscsi_handle_crq: - Handles and frees received events in the CRQ
1757 * @crq:	Command/Response queue
1758 * @hostdata:	ibmvscsi_host_data of host
1759 *
1760*/
1761static void ibmvscsi_handle_crq(struct viosrp_crq *crq,
1762				struct ibmvscsi_host_data *hostdata)
1763{
1764	long rc;
1765	unsigned long flags;
1766	/* The hypervisor copies our tag value here so no byteswapping */
1767	struct srp_event_struct *evt_struct =
1768			(__force struct srp_event_struct *)crq->IU_data_ptr;
1769	switch (crq->valid) {
1770	case 0xC0:		/* initialization */
1771		switch (crq->format) {
1772		case 0x01:	/* Initialization message */
1773			dev_info(hostdata->dev, "partner initialized\n");
1774			/* Send back a response */
1775			rc = ibmvscsi_send_crq(hostdata, 0xC002000000000000LL, 0);
1776			if (rc == 0) {
1777				/* Now login */
1778				init_adapter(hostdata);
1779			} else {
1780				dev_err(hostdata->dev, "Unable to send init rsp. rc=%ld\n", rc);
1781			}
1782
1783			break;
1784		case 0x02:	/* Initialization response */
1785			dev_info(hostdata->dev, "partner initialization complete\n");
1786
1787			/* Now login */
1788			init_adapter(hostdata);
1789			break;
1790		default:
1791			dev_err(hostdata->dev, "unknown crq message type: %d\n", crq->format);
1792		}
1793		return;
1794	case 0xFF:	/* Hypervisor telling us the connection is closed */
1795		scsi_block_requests(hostdata->host);
1796		atomic_set(&hostdata->request_limit, 0);
1797		if (crq->format == 0x06) {
1798			/* We need to re-setup the interpartition connection */
1799			dev_info(hostdata->dev, "Re-enabling adapter!\n");
1800			hostdata->client_migrated = 1;
1801			hostdata->reenable_crq = 1;
1802			purge_requests(hostdata, DID_REQUEUE);
1803			wake_up(&hostdata->work_wait_q);
1804		} else {
1805			dev_err(hostdata->dev, "Virtual adapter failed rc %d!\n",
1806				crq->format);
1807			ibmvscsi_reset_host(hostdata);
1808		}
1809		return;
1810	case 0x80:		/* real payload */
1811		break;
1812	default:
1813		dev_err(hostdata->dev, "got an invalid message type 0x%02x\n",
1814			crq->valid);
1815		return;
1816	}
1817
1818	/* The only kind of payload CRQs we should get are responses to
1819	 * things we send. Make sure this response is to something we
1820	 * actually sent
1821	 */
1822	if (!valid_event_struct(&hostdata->pool, evt_struct)) {
1823		dev_err(hostdata->dev, "returned correlation_token 0x%p is invalid!\n",
1824		       evt_struct);
1825		return;
1826	}
1827
1828	if (atomic_read(&evt_struct->free)) {
1829		dev_err(hostdata->dev, "received duplicate correlation_token 0x%p!\n",
1830			evt_struct);
1831		return;
1832	}
1833
1834	if (crq->format == VIOSRP_SRP_FORMAT)
1835		atomic_add(be32_to_cpu(evt_struct->xfer_iu->srp.rsp.req_lim_delta),
1836			   &hostdata->request_limit);
1837
1838	del_timer(&evt_struct->timer);
1839
1840	if ((crq->status != VIOSRP_OK && crq->status != VIOSRP_OK2) && evt_struct->cmnd)
1841		evt_struct->cmnd->result = DID_ERROR << 16;
1842	if (evt_struct->done)
1843		evt_struct->done(evt_struct);
1844	else
1845		dev_err(hostdata->dev, "returned done() is NULL; not running it!\n");
1846
1847	/*
1848	 * Lock the host_lock before messing with these structures, since we
1849	 * are running in a task context
1850	 */
1851	spin_lock_irqsave(evt_struct->hostdata->host->host_lock, flags);
1852	list_del(&evt_struct->list);
1853	free_event_struct(&evt_struct->hostdata->pool, evt_struct);
1854	spin_unlock_irqrestore(evt_struct->hostdata->host->host_lock, flags);
1855}
1856
1857/**
1858 * ibmvscsi_get_host_config: Send the command to the server to get host
1859 * configuration data.  The data is opaque to us.
1860 */
1861static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
1862				   unsigned char *buffer, int length)
1863{
1864	struct viosrp_host_config *host_config;
1865	struct srp_event_struct *evt_struct;
1866	unsigned long flags;
1867	dma_addr_t addr;
1868	int rc;
1869
1870	evt_struct = get_event_struct(&hostdata->pool);
1871	if (!evt_struct) {
1872		dev_err(hostdata->dev, "couldn't allocate event for HOST_CONFIG!\n");
1873		return -1;
1874	}
1875
1876	init_event_struct(evt_struct,
1877			  sync_completion,
1878			  VIOSRP_MAD_FORMAT,
1879			  info_timeout);
1880
1881	host_config = &evt_struct->iu.mad.host_config;
1882
1883	/* The transport length field is only 16-bit */
1884	length = min(0xffff, length);
1885
1886	/* Set up a lun reset SRP command */
1887	memset(host_config, 0x00, sizeof(*host_config));
1888	host_config->common.type = cpu_to_be32(VIOSRP_HOST_CONFIG_TYPE);
1889	host_config->common.length = cpu_to_be16(length);
1890	addr = dma_map_single(hostdata->dev, buffer, length, DMA_BIDIRECTIONAL);
1891
1892	if (dma_mapping_error(hostdata->dev, addr)) {
1893		if (!firmware_has_feature(FW_FEATURE_CMO))
1894			dev_err(hostdata->dev,
1895			        "dma_mapping error getting host config\n");
1896		free_event_struct(&hostdata->pool, evt_struct);
1897		return -1;
1898	}
1899
1900	host_config->buffer = cpu_to_be64(addr);
1901
1902	init_completion(&evt_struct->comp);
1903	spin_lock_irqsave(hostdata->host->host_lock, flags);
1904	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, info_timeout * 2);
1905	spin_unlock_irqrestore(hostdata->host->host_lock, flags);
1906	if (rc == 0)
1907		wait_for_completion(&evt_struct->comp);
1908	dma_unmap_single(hostdata->dev, addr, length, DMA_BIDIRECTIONAL);
1909
1910	return rc;
1911}
1912
1913/**
1914 * ibmvscsi_slave_configure: Set the "allow_restart" flag for each disk.
1915 * @sdev:	struct scsi_device device to configure
1916 *
1917 * Enable allow_restart for a device if it is a disk.  Adjust the
1918 * queue_depth here also as is required by the documentation for
1919 * struct scsi_host_template.
1920 */
1921static int ibmvscsi_slave_configure(struct scsi_device *sdev)
1922{
1923	struct Scsi_Host *shost = sdev->host;
1924	unsigned long lock_flags = 0;
1925
1926	spin_lock_irqsave(shost->host_lock, lock_flags);
1927	if (sdev->type == TYPE_DISK) {
1928		sdev->allow_restart = 1;
1929		blk_queue_rq_timeout(sdev->request_queue, 120 * HZ);
1930	}
1931	spin_unlock_irqrestore(shost->host_lock, lock_flags);
1932	return 0;
1933}
1934
1935/**
1936 * ibmvscsi_change_queue_depth - Change the device's queue depth
1937 * @sdev:	scsi device struct
1938 * @qdepth:	depth to set
1939 * @reason:	calling context
1940 *
1941 * Return value:
1942 * 	actual depth set
1943 **/
1944static int ibmvscsi_change_queue_depth(struct scsi_device *sdev, int qdepth)
1945{
1946	if (qdepth > IBMVSCSI_MAX_CMDS_PER_LUN)
1947		qdepth = IBMVSCSI_MAX_CMDS_PER_LUN;
1948	return scsi_change_queue_depth(sdev, qdepth);
1949}
1950
1951/* ------------------------------------------------------------
1952 * sysfs attributes
1953 */
1954static ssize_t show_host_vhost_loc(struct device *dev,
1955				   struct device_attribute *attr, char *buf)
1956{
1957	struct Scsi_Host *shost = class_to_shost(dev);
1958	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1959	int len;
1960
1961	len = snprintf(buf, sizeof(hostdata->caps.loc), "%s\n",
1962		       hostdata->caps.loc);
1963	return len;
1964}
1965
1966static struct device_attribute ibmvscsi_host_vhost_loc = {
1967	.attr = {
1968		 .name = "vhost_loc",
1969		 .mode = S_IRUGO,
1970		 },
1971	.show = show_host_vhost_loc,
1972};
1973
1974static ssize_t show_host_vhost_name(struct device *dev,
1975				    struct device_attribute *attr, char *buf)
1976{
1977	struct Scsi_Host *shost = class_to_shost(dev);
1978	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1979	int len;
1980
1981	len = snprintf(buf, sizeof(hostdata->caps.name), "%s\n",
1982		       hostdata->caps.name);
1983	return len;
1984}
1985
1986static struct device_attribute ibmvscsi_host_vhost_name = {
1987	.attr = {
1988		 .name = "vhost_name",
1989		 .mode = S_IRUGO,
1990		 },
1991	.show = show_host_vhost_name,
1992};
1993
1994static ssize_t show_host_srp_version(struct device *dev,
1995				     struct device_attribute *attr, char *buf)
1996{
1997	struct Scsi_Host *shost = class_to_shost(dev);
1998	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
1999	int len;
2000
2001	len = snprintf(buf, PAGE_SIZE, "%s\n",
2002		       hostdata->madapter_info.srp_version);
2003	return len;
2004}
2005
2006static struct device_attribute ibmvscsi_host_srp_version = {
2007	.attr = {
2008		 .name = "srp_version",
2009		 .mode = S_IRUGO,
2010		 },
2011	.show = show_host_srp_version,
2012};
2013
2014static ssize_t show_host_partition_name(struct device *dev,
2015					struct device_attribute *attr,
2016					char *buf)
2017{
2018	struct Scsi_Host *shost = class_to_shost(dev);
2019	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2020	int len;
2021
2022	len = snprintf(buf, PAGE_SIZE, "%s\n",
2023		       hostdata->madapter_info.partition_name);
2024	return len;
2025}
2026
2027static struct device_attribute ibmvscsi_host_partition_name = {
2028	.attr = {
2029		 .name = "partition_name",
2030		 .mode = S_IRUGO,
2031		 },
2032	.show = show_host_partition_name,
2033};
2034
2035static ssize_t show_host_partition_number(struct device *dev,
2036					  struct device_attribute *attr,
2037					  char *buf)
2038{
2039	struct Scsi_Host *shost = class_to_shost(dev);
2040	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2041	int len;
2042
2043	len = snprintf(buf, PAGE_SIZE, "%d\n",
2044		       hostdata->madapter_info.partition_number);
2045	return len;
2046}
2047
2048static struct device_attribute ibmvscsi_host_partition_number = {
2049	.attr = {
2050		 .name = "partition_number",
2051		 .mode = S_IRUGO,
2052		 },
2053	.show = show_host_partition_number,
2054};
2055
2056static ssize_t show_host_mad_version(struct device *dev,
2057				     struct device_attribute *attr, char *buf)
2058{
2059	struct Scsi_Host *shost = class_to_shost(dev);
2060	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2061	int len;
2062
2063	len = snprintf(buf, PAGE_SIZE, "%d\n",
2064		       hostdata->madapter_info.mad_version);
2065	return len;
2066}
2067
2068static struct device_attribute ibmvscsi_host_mad_version = {
2069	.attr = {
2070		 .name = "mad_version",
2071		 .mode = S_IRUGO,
2072		 },
2073	.show = show_host_mad_version,
2074};
2075
2076static ssize_t show_host_os_type(struct device *dev,
2077				 struct device_attribute *attr, char *buf)
2078{
2079	struct Scsi_Host *shost = class_to_shost(dev);
2080	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2081	int len;
2082
2083	len = snprintf(buf, PAGE_SIZE, "%d\n", hostdata->madapter_info.os_type);
2084	return len;
2085}
2086
2087static struct device_attribute ibmvscsi_host_os_type = {
2088	.attr = {
2089		 .name = "os_type",
2090		 .mode = S_IRUGO,
2091		 },
2092	.show = show_host_os_type,
2093};
2094
2095static ssize_t show_host_config(struct device *dev,
2096				struct device_attribute *attr, char *buf)
2097{
2098	struct Scsi_Host *shost = class_to_shost(dev);
2099	struct ibmvscsi_host_data *hostdata = shost_priv(shost);
2100
2101	/* returns null-terminated host config data */
2102	if (ibmvscsi_do_host_config(hostdata, buf, PAGE_SIZE) == 0)
2103		return strlen(buf);
2104	else
2105		return 0;
2106}
2107
2108static struct device_attribute ibmvscsi_host_config = {
2109	.attr = {
2110		 .name = "config",
2111		 .mode = S_IRUGO,
2112		 },
2113	.show = show_host_config,
2114};
2115
2116static struct device_attribute *ibmvscsi_attrs[] = {
2117	&ibmvscsi_host_vhost_loc,
2118	&ibmvscsi_host_vhost_name,
2119	&ibmvscsi_host_srp_version,
2120	&ibmvscsi_host_partition_name,
2121	&ibmvscsi_host_partition_number,
2122	&ibmvscsi_host_mad_version,
2123	&ibmvscsi_host_os_type,
2124	&ibmvscsi_host_config,
2125	NULL
2126};
2127
2128/* ------------------------------------------------------------
2129 * SCSI driver registration
2130 */
2131static struct scsi_host_template driver_template = {
2132	.module = THIS_MODULE,
2133	.name = "IBM POWER Virtual SCSI Adapter " IBMVSCSI_VERSION,
2134	.proc_name = "ibmvscsi",
2135	.queuecommand = ibmvscsi_queuecommand,
2136	.eh_abort_handler = ibmvscsi_eh_abort_handler,
2137	.eh_device_reset_handler = ibmvscsi_eh_device_reset_handler,
2138	.eh_host_reset_handler = ibmvscsi_eh_host_reset_handler,
2139	.slave_configure = ibmvscsi_slave_configure,
2140	.change_queue_depth = ibmvscsi_change_queue_depth,
2141	.cmd_per_lun = IBMVSCSI_CMDS_PER_LUN_DEFAULT,
2142	.can_queue = IBMVSCSI_MAX_REQUESTS_DEFAULT,
2143	.this_id = -1,
2144	.sg_tablesize = SG_ALL,
2145	.use_clustering = ENABLE_CLUSTERING,
2146	.shost_attrs = ibmvscsi_attrs,
2147};
2148
2149/**
2150 * ibmvscsi_get_desired_dma - Calculate IO memory desired by the driver
2151 *
2152 * @vdev: struct vio_dev for the device whose desired IO mem is to be returned
2153 *
2154 * Return value:
2155 *	Number of bytes of IO data the driver will need to perform well.
2156 */
2157static unsigned long ibmvscsi_get_desired_dma(struct vio_dev *vdev)
2158{
2159	/* iu_storage data allocated in initialize_event_pool */
2160	unsigned long desired_io = max_events * sizeof(union viosrp_iu);
2161
2162	/* add io space for sg data */
2163	desired_io += (IBMVSCSI_MAX_SECTORS_DEFAULT * 512 *
2164	                     IBMVSCSI_CMDS_PER_LUN_DEFAULT);
2165
2166	return desired_io;
2167}
2168
2169static void ibmvscsi_do_work(struct ibmvscsi_host_data *hostdata)
2170{
2171	int rc;
2172	char *action = "reset";
2173
2174	if (hostdata->reset_crq) {
2175		smp_rmb();
2176		hostdata->reset_crq = 0;
2177
2178		rc = ibmvscsi_reset_crq_queue(&hostdata->queue, hostdata);
2179		if (!rc)
2180			rc = ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0);
2181		vio_enable_interrupts(to_vio_dev(hostdata->dev));
2182	} else if (hostdata->reenable_crq) {
2183		smp_rmb();
2184		action = "enable";
2185		rc = ibmvscsi_reenable_crq_queue(&hostdata->queue, hostdata);
2186		hostdata->reenable_crq = 0;
2187		if (!rc)
2188			rc = ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0);
2189	} else
2190		return;
2191
2192	if (rc) {
2193		atomic_set(&hostdata->request_limit, -1);
2194		dev_err(hostdata->dev, "error after %s\n", action);
2195	}
2196
2197	scsi_unblock_requests(hostdata->host);
2198}
2199
2200static int ibmvscsi_work_to_do(struct ibmvscsi_host_data *hostdata)
2201{
2202	if (kthread_should_stop())
2203		return 1;
2204	else if (hostdata->reset_crq) {
2205		smp_rmb();
2206		return 1;
2207	} else if (hostdata->reenable_crq) {
2208		smp_rmb();
2209		return 1;
2210	}
2211
2212	return 0;
2213}
2214
2215static int ibmvscsi_work(void *data)
2216{
2217	struct ibmvscsi_host_data *hostdata = data;
2218	int rc;
2219
2220	set_user_nice(current, MIN_NICE);
2221
2222	while (1) {
2223		rc = wait_event_interruptible(hostdata->work_wait_q,
2224					      ibmvscsi_work_to_do(hostdata));
2225
2226		BUG_ON(rc);
2227
2228		if (kthread_should_stop())
2229			break;
2230
2231		ibmvscsi_do_work(hostdata);
2232	}
2233
2234	return 0;
2235}
2236
2237/**
2238 * Called by bus code for each adapter
2239 */
2240static int ibmvscsi_probe(struct vio_dev *vdev, const struct vio_device_id *id)
2241{
2242	struct ibmvscsi_host_data *hostdata;
2243	struct Scsi_Host *host;
2244	struct device *dev = &vdev->dev;
2245	struct srp_rport_identifiers ids;
2246	struct srp_rport *rport;
2247	unsigned long wait_switch = 0;
2248	int rc;
2249
2250	dev_set_drvdata(&vdev->dev, NULL);
2251
2252	host = scsi_host_alloc(&driver_template, sizeof(*hostdata));
2253	if (!host) {
2254		dev_err(&vdev->dev, "couldn't allocate host data\n");
2255		goto scsi_host_alloc_failed;
2256	}
2257
2258	host->transportt = ibmvscsi_transport_template;
2259	hostdata = shost_priv(host);
2260	memset(hostdata, 0x00, sizeof(*hostdata));
2261	INIT_LIST_HEAD(&hostdata->sent);
2262	init_waitqueue_head(&hostdata->work_wait_q);
2263	hostdata->host = host;
2264	hostdata->dev = dev;
2265	atomic_set(&hostdata->request_limit, -1);
2266	hostdata->host->max_sectors = IBMVSCSI_MAX_SECTORS_DEFAULT;
2267
2268	if (map_persist_bufs(hostdata)) {
2269		dev_err(&vdev->dev, "couldn't map persistent buffers\n");
2270		goto persist_bufs_failed;
2271	}
2272
2273	hostdata->work_thread = kthread_run(ibmvscsi_work, hostdata, "%s_%d",
2274					    "ibmvscsi", host->host_no);
2275
2276	if (IS_ERR(hostdata->work_thread)) {
2277		dev_err(&vdev->dev, "couldn't initialize kthread. rc=%ld\n",
2278			PTR_ERR(hostdata->work_thread));
2279		goto init_crq_failed;
2280	}
2281
2282	rc = ibmvscsi_init_crq_queue(&hostdata->queue, hostdata, max_events);
2283	if (rc != 0 && rc != H_RESOURCE) {
2284		dev_err(&vdev->dev, "couldn't initialize crq. rc=%d\n", rc);
2285		goto kill_kthread;
2286	}
2287	if (initialize_event_pool(&hostdata->pool, max_events, hostdata) != 0) {
2288		dev_err(&vdev->dev, "couldn't initialize event pool\n");
2289		goto init_pool_failed;
2290	}
2291
2292	host->max_lun = 8;
2293	host->max_id = max_id;
2294	host->max_channel = max_channel;
2295	host->max_cmd_len = 16;
2296
2297	if (scsi_add_host(hostdata->host, hostdata->dev))
2298		goto add_host_failed;
2299
2300	/* we don't have a proper target_port_id so let's use the fake one */
2301	memcpy(ids.port_id, hostdata->madapter_info.partition_name,
2302	       sizeof(ids.port_id));
2303	ids.roles = SRP_RPORT_ROLE_TARGET;
2304	rport = srp_rport_add(host, &ids);
2305	if (IS_ERR(rport))
2306		goto add_srp_port_failed;
2307
2308	/* Try to send an initialization message.  Note that this is allowed
2309	 * to fail if the other end is not acive.  In that case we don't
2310	 * want to scan
2311	 */
2312	if (ibmvscsi_send_crq(hostdata, 0xC001000000000000LL, 0) == 0
2313	    || rc == H_RESOURCE) {
2314		/*
2315		 * Wait around max init_timeout secs for the adapter to finish
2316		 * initializing. When we are done initializing, we will have a
2317		 * valid request_limit.  We don't want Linux scanning before
2318		 * we are ready.
2319		 */
2320		for (wait_switch = jiffies + (init_timeout * HZ);
2321		     time_before(jiffies, wait_switch) &&
2322		     atomic_read(&hostdata->request_limit) < 2;) {
2323
2324			msleep(10);
2325		}
2326
2327		/* if we now have a valid request_limit, initiate a scan */
2328		if (atomic_read(&hostdata->request_limit) > 0)
2329			scsi_scan_host(host);
2330	}
2331
2332	dev_set_drvdata(&vdev->dev, hostdata);
2333	return 0;
2334
2335      add_srp_port_failed:
2336	scsi_remove_host(hostdata->host);
2337      add_host_failed:
2338	release_event_pool(&hostdata->pool, hostdata);
2339      init_pool_failed:
2340	ibmvscsi_release_crq_queue(&hostdata->queue, hostdata, max_events);
2341      kill_kthread:
2342      kthread_stop(hostdata->work_thread);
2343      init_crq_failed:
2344	unmap_persist_bufs(hostdata);
2345      persist_bufs_failed:
2346	scsi_host_put(host);
2347      scsi_host_alloc_failed:
2348	return -1;
2349}
2350
2351static int ibmvscsi_remove(struct vio_dev *vdev)
2352{
2353	struct ibmvscsi_host_data *hostdata = dev_get_drvdata(&vdev->dev);
2354	unmap_persist_bufs(hostdata);
2355	release_event_pool(&hostdata->pool, hostdata);
2356	ibmvscsi_release_crq_queue(&hostdata->queue, hostdata,
2357					max_events);
2358
2359	kthread_stop(hostdata->work_thread);
2360	srp_remove_host(hostdata->host);
2361	scsi_remove_host(hostdata->host);
2362	scsi_host_put(hostdata->host);
2363
2364	return 0;
2365}
2366
2367/**
2368 * ibmvscsi_resume: Resume from suspend
2369 * @dev:	device struct
2370 *
2371 * We may have lost an interrupt across suspend/resume, so kick the
2372 * interrupt handler
2373 */
2374static int ibmvscsi_resume(struct device *dev)
2375{
2376	struct ibmvscsi_host_data *hostdata = dev_get_drvdata(dev);
2377	vio_disable_interrupts(to_vio_dev(hostdata->dev));
2378	tasklet_schedule(&hostdata->srp_task);
2379
2380	return 0;
2381}
2382
2383/**
2384 * ibmvscsi_device_table: Used by vio.c to match devices in the device tree we
2385 * support.
2386 */
2387static struct vio_device_id ibmvscsi_device_table[] = {
2388	{"vscsi", "IBM,v-scsi"},
2389	{ "", "" }
2390};
2391MODULE_DEVICE_TABLE(vio, ibmvscsi_device_table);
2392
2393static struct dev_pm_ops ibmvscsi_pm_ops = {
2394	.resume = ibmvscsi_resume
2395};
2396
2397static struct vio_driver ibmvscsi_driver = {
2398	.id_table = ibmvscsi_device_table,
2399	.probe = ibmvscsi_probe,
2400	.remove = ibmvscsi_remove,
2401	.get_desired_dma = ibmvscsi_get_desired_dma,
2402	.name = "ibmvscsi",
2403	.pm = &ibmvscsi_pm_ops,
2404};
2405
2406static struct srp_function_template ibmvscsi_transport_functions = {
2407};
2408
2409int __init ibmvscsi_module_init(void)
2410{
2411	int ret;
2412
2413	/* Ensure we have two requests to do error recovery */
2414	driver_template.can_queue = max_requests;
2415	max_events = max_requests + 2;
2416
2417	if (!firmware_has_feature(FW_FEATURE_VIO))
2418		return -ENODEV;
2419
2420	ibmvscsi_transport_template =
2421		srp_attach_transport(&ibmvscsi_transport_functions);
2422	if (!ibmvscsi_transport_template)
2423		return -ENOMEM;
2424
2425	ret = vio_register_driver(&ibmvscsi_driver);
2426	if (ret)
2427		srp_release_transport(ibmvscsi_transport_template);
2428	return ret;
2429}
2430
2431void __exit ibmvscsi_module_exit(void)
2432{
2433	vio_unregister_driver(&ibmvscsi_driver);
2434	srp_release_transport(ibmvscsi_transport_template);
2435}
2436
2437module_init(ibmvscsi_module_init);
2438module_exit(ibmvscsi_module_exit);
2439